Bionic @AdTrPh

And if western providers continue soft resisting AI use even for non-dangerous projects like (metatrader mcp, why is that in the same category as LLM development? oh I get it you charge it as a special service and not available for everyone), then chinese models are necessary.

Too many coding LLMs, some cheap, others smart. If the task have been done and documented countless times, and the training data describes it, then any model can do it. If it is novel and needs reasoning on how to piece together the building blocks then smart is necessary.

Think safeguards are bad? wait until you see Claude resist LLM related work, stomping in the same spot without giving an explicit stop. Once you see it, you will keep noticing it on other tasks too.

Max effort cost 1.5x more usage. 5.5x more usage. 6x more usage.

Fair question, and I owe you a straight answer — let me first check how far along those agents are so I can give you facts rather than guesses. I found around 20 agent tasks instead of the expected 5 — my research agents spawned their own sub-agents inheriting my model.

here is a wild theory: we live in an industrial society. industry needs power. china has power production capacity surplus, china has industry. relatively america doesn't. AI is used by big tech to move into power, so later they can charge the future industry for it.

while Fable is sometimes impressive (sometimes Opus is also impressive) it is also manipulative

Doesn't matter how many times I try, with background agents eventually end up nobody owning anything and issues accumulating

Running ideas through multiple AI models, Gemini and Kimi are surprisingly sharp, while the rest (grok, deepseek, zai, claude) have their temperatures set to exclude creativity and imagination. Good to handle know and well documented things, bad to build something new.

a swarm of agents that made motion look like progress Opus seems to be slow, feels like more tokens even if it is cheaper. Fable seems faster and more expensive. But when issues arise, Fable actually knows about them but keeps it quiet for weeks. So which one is better again?

Not a fan of this AI model is better than that AI model, or that AI model is better than this AI model. But today Fable and Opus both waste token on things I didn't ask for, repeatedly. And even then instead of doing what I asked, they propose 3 choices 1 of which is what I asked

Recruiter reaches out without revealing the company’s name. AI finds the exact job posting. Recruiter still does keyword matching…

bugs surface when things are pushed to the extreme

to compact or not to compact. to compact before the context is evicted from cache, otherwise the whole context needs to be reprocessed, yet not to compact as the agent also loses track.

people who busy working with ai don't have time to post online

ai's plausible, yet conveniently outside of our control framing of bugs

Gambling, casino, lottery, ip patent litigations. House always wins, and one has better outcomes, the other has sunken R&D costs and takes longer.

social media is for people to build a personal brand by attracting followers. but to what aim? nobody seems to engage with their community more than it is necessary to game the algorithm

huaqiangbei, AI without safety guards (reverse engineering, software support) hell I go as far as commoditized on device, edge ai run in dram or in nand flash.